xref: /llvm-project/llvm/test/CodeGen/ARM/intrinsics-overflow.ll (revision 2ac116291729fa8f9af82f4b7005c0839aff7fa9)
1; RUN: llc < %s -mtriple=arm-linux -mcpu=generic -verify-machineinstrs | FileCheck %s --check-prefix=CHECK --check-prefix=ARM
2; RUN: llc < %s -mtriple=thumbv6m-eabi -verify-machineinstrs | FileCheck %s --check-prefix=CHECK --check-prefix=THUMBV6
3; RUN: llc < %s -mtriple=thumbv7-eabi -verify-machineinstrs | FileCheck %s --check-prefix=CHECK --check-prefix=THUMBV7
4
5define i32 @uadd_overflow(i32 %a, i32 %b) #0 {
6  %sadd = tail call { i32, i1 } @llvm.uadd.with.overflow.i32(i32 %a, i32 %b)
7  %1 = extractvalue { i32, i1 } %sadd, 1
8  %2 = zext i1 %1 to i32
9  ret i32 %2
10
11  ; CHECK-LABEL: uadd_overflow:
12
13  ; ARM: adds r[[R0:[0-9]+]], r[[R0]], r[[R1:[0-9]+]]
14  ; ARM: mov r[[R2:[0-9]+]], #0
15  ; ARM: adc r[[R0]], r[[R2]], #0
16
17  ; THUMBV6: movs    r[[R2:[0-9]+]], #0
18  ; THUMBV6: adds    r[[R0:[0-9]+]], r[[R0]], r[[R1:[0-9]+]]
19  ; THUMBV6: adcs    r[[R2]], r[[R2]]
20  ; THUMBV6: mov     r[[R0]], r[[R2]]
21
22  ; THUMBV7: adds  r[[R0:[0-9]+]], r[[R0]], r[[R1:[0-9]+]]
23  ; THUMBV7: mov.w r[[R2:[0-9]+]], #0
24  ; THUMBV7: adc   r[[R0]], r[[R2]], #0
25}
26
27
28define i32 @sadd_overflow(i32 %a, i32 %b) #0 {
29  %sadd = tail call { i32, i1 } @llvm.sadd.with.overflow.i32(i32 %a, i32 %b)
30  %1 = extractvalue { i32, i1 } %sadd, 1
31  %2 = zext i1 %1 to i32
32  ret i32 %2
33
34  ; CHECK-LABEL: sadd_overflow:
35
36  ; ARM: adds r[[R2:[0-9]+]], r[[R0:[0-9]+]], r[[R1:[0-9]+]]
37  ; ARM: mov r[[R0]], #1
38  ; ARM: movvc r[[R0]], #0
39  ; ARM: mov pc, lr
40
41  ; THUMBV6: adds    r1, r0, r1
42  ; THUMBV6: cmp     r1, r0
43  ; THUMBV6: bvc     .LBB1_2
44
45  ; THUMBV7: adds  r[[R2:[0-9]+]], r[[R0]], r[[R1:[0-9]+]]
46  ; THUMBV7: mov.w r[[R0:[0-9]+]], #1
47  ; THUMBV7: it    vc
48  ; THUMBV7: movvc r[[R0]], #0
49}
50
51define i32 @usub_overflow(i32 %a, i32 %b) #0 {
52  %sadd = tail call { i32, i1 } @llvm.usub.with.overflow.i32(i32 %a, i32 %b)
53  %1 = extractvalue { i32, i1 } %sadd, 1
54  %2 = zext i1 %1 to i32
55  ret i32 %2
56
57  ; CHECK-LABEL: usub_overflow:
58
59  ; ARM: subs    r[[R0:[0-9]+]], r[[R0]], r[[R1:[0-9]+]]
60  ; ARM: mov     r[[R2:[0-9]+]], #0
61  ; ARM: adc     r[[R0]], r[[R2]], #0
62  ; ARM: rsb     r[[R0]], r[[R0]], #1
63
64  ; THUMBV6: movs    r[[R2:[0-9]+]], #0
65  ; THUMBV6: subs    r[[R0:[0-9]+]], r[[R0]], r[[R1:[0-9]+]]
66  ; THUMBV6: adcs    r[[R2]], r[[R2]]
67  ; THUMBV6: movs    r[[R0]], #1
68  ; THUMBV6: subs    r[[R0]], r[[R0]], r[[R2]]
69
70  ; THUMBV7: subs    r[[R0:[0-9]+]], r[[R0]], r[[R1:[0-9]+]]
71  ; THUMBV7: mov.w   r[[R2:[0-9]+]], #0
72  ; THUMBV7: adc     r[[R0]], r[[R2]], #0
73  ; THUMBV7: rsb.w   r[[R0]], r[[R0]], #1
74
75  ; We should know that the overflow is just 1 bit,
76  ; no need to clear any other bit
77  ; CHECK-NOT: and
78}
79
80define i32 @ssub_overflow(i32 %a, i32 %b) #0 {
81  %sadd = tail call { i32, i1 } @llvm.ssub.with.overflow.i32(i32 %a, i32 %b)
82  %1 = extractvalue { i32, i1 } %sadd, 1
83  %2 = zext i1 %1 to i32
84  ret i32 %2
85
86  ; CHECK-LABEL: ssub_overflow:
87
88  ; ARM: mov r[[R2]], #1
89  ; ARM: cmp r[[R0]], r[[R1]]
90  ; ARM: movvc r[[R2]], #0
91
92  ; THUMBV6: cmp     r0, r1
93  ; THUMBV6: bvc     .LBB3_2
94
95  ; THUMBV7: movs  r[[R2:[0-9]+]], #1
96  ; THUMBV7: cmp   r[[R0:[0-9]+]], r[[R1:[0-9]+]]
97  ; THUMBV7: it    vc
98  ; THUMBV7: movvc r[[R2]], #0
99  ; THUMBV7: mov   r[[R0]], r[[R2]]
100}
101
102declare { i32, i1 } @llvm.uadd.with.overflow.i32(i32, i32) #1
103declare { i32, i1 } @llvm.sadd.with.overflow.i32(i32, i32) #2
104declare { i32, i1 } @llvm.usub.with.overflow.i32(i32, i32) #3
105declare { i32, i1 } @llvm.ssub.with.overflow.i32(i32, i32) #4
106